Figure 2.

Characterizing Dependence on Spectral Cues in a Model of Sound Localization

(A) Stimuli with randomized spectra (gray). Amplitude varied considerably across frequency on a single trial (black), but the mean spectrum across trials was relatively flat (red).

(B) On each trial, a stimulus was presented at one of 12 virtual locations, which was achieved by convolving the source stimulus with the external ear’s transfer function appropriate for that location. Examples of flat (blue) and randomized-spectrum (black) stimuli convolved with the ear’s transfer function (DTF) for 0° azimuth.

(C) Directional transfer function (DTF) used by the monaural model to infer sound location, derived from acoustical measurements from a single ferret. This illustrates the spectral localization cues generated in the horizontal plane by the immobile right external ear of this individual. Warmer colors indicate higher gains.

(D) In order to determine its response, the model computed the correlation between the spectrum of the spatially filtered stimulus and the ear’s transfer function associated with each of the 12 possible locations (C), identified the transfer function that was maximally correlated with the stimulus, and selected the corresponding location as its response (denoted by arrows). The correlation between the amplitude spectra of stimuli corresponding to a virtual azimuth of 0° and the transfer functions associated with different locations is shown for stimuli with flat (blue) and randomized (black) source spectra.

(E) The mean spectrum associated with each response location (gray/black line) typically differed from the overall mean across all locations (red). Data are shown for trials on which the model responded to a stimulus location of 60°.

(F) Reverse correlation map (RCM) constructed by thresholding (±1.5 SD) the mean spectra shown in (E). Colors show differences between the overall mean and the mean stimulus associated with each response location.

(G) Degree of similarity between the RCM in (F) and the gain of the DTF shown in (C), as determined by applying linear regression to the point-wise comparison between the two measures. Each dot shows a particular combination of frequency and location. Data corresponding to high (>8 kHz; black) and low (<8 kHz; gray) frequencies are plotted separately.

(H) RCM obtained using a binaural model.